Frailty Assessment in Hospitalized Older Adults Using the Electronic Health Record

Abstract

Introduction:

Frailty, a clinical syndrome of decreased physiologic reserve and dysregulation in multiple physiologic systems, is associated with increased risk for adverse outcomes.

Purpose:

The aim of this retrospective, cross-sectional, correlational study was to characterize frailty in older adults admitted to a tertiary-care hospital using a biopsychosocial frailty assessment and to determine associations between frailty and time to in-hospital mortality and 30-day rehospitalization.

Methods:

The sample included 278 patients ≥55 years old admitted to medicine units. Frailty was determined using clinical data from the electronic health record (EHR) for symptoms, syndromes, and conditions and laboratory data for four serum biomarkers. A frailty risk score (FRS) was created from 16 risk factors, and relationships between the FRS and outcomes were examined.

Results:

The mean age of the sample was 70.2 years and mean FRS was 9.4 (SD, 2.2). Increased FRS was significantly associated with increased risk of death (hazard ratio = 1.77−2.27 for 3 days ≤ length of stay (LOS) ≤7 days), but depended upon LOS (p < .001). Frailty was marginally associated with rehospitalization for those who did not die in hospital (adjusted odds ratio = 1.18, p = .086, area under the curve [AUC] = 0.66, 95% confidence interval for AUC = [0.57, 0.76]).

Discussion:

Clinical data in the EHR can be used for frailty assessment. Informatics may facilitate data aggregation and decision support. Because frailty is potentially preventable and treatable, early detection is crucial to delivery of tailored interventions and optimal patient outcomes.

The aging population is growing at a staggering rate worldwide. Along with the rapid increase in the number and longevity of older adults, there is an expected increase in the prevalence of frailty. Although many older adults experience some decline in physical, cognitive, and/or functional status at advanced ages, not all are destined to become frail. Frailty is a clinical syndrome characterized by multifactorial etiologies, vague symptom presentation, and vulnerability to adverse health outcomes that is disproportionate to the instigating stressors. There is evidence that frail older adults are more susceptible to the stresses of acute illness and hospitalization, are at greater risk for complications from treatment and surgery, and experience worse outcomes than persons who are not frail (Afilalo et al., 2014; Basic & Shanley, 2015; N. M. de Vries, Staal, van Ravensberg, Hobbelen, & Olde Rikkert, 2011). Research indicates that frailty assessment is a consistent and significant predictor of deleterious outcomes in hospitalized older adults (Afilalo et al., 2014; Hilmer et al., 2009). However, though there is widespread agreement that frailty is an important clinical phenomenon, the literature is conflicted about how to best assess frailty in this population (Sternberg, Wershof Schwartz, Karunananthan, Bergman, & Mark Clarfield, 2011). Assessment tools are thus needed to identify frail patients as well as to guide clinical decision-making and care management and to monitor treatment effects.

Frailty is a clinical syndrome resulting from physiologic impairments and failed integrative responses to stressors in multiple interrelated systems and is distinguished by reduced resilience and poor bounce-back from adverse health events (Cesari, 2011; Sternberg et al., 2011; Yao, Li, & Leng, 2011). Frailty differs from normal aging in its clinical manifestations and represents the combined effects on health of aging, disease, and functional, psychosocial, and environmental factors. Many investigators consider frailty as an intermediate state that falls on a continuum, with normal aging, resilience, and prefrail states on one end and severe frailty, disability, and end of life on the other (Dramé et al., 2008). Also, frailty is a dynamic state, in that level of frailty may fluctuate, especially in its early phases, which implies that there is an intervention window for prevention or reversal of frailty. However, in later stages, frailty is more intractable to interventions due to advanced multisystem dysregulation. Thus, early identification is crucial to prevent or halt progression of frailty and maximize function and quality of life. Early identification is particularly important in hospitalized older adults, for whom the negative impact of acute illness on health status can be magnified in the context of frailty and return to baseline function is less likely.

Hospitalization is hazardous for older people. Many hospitalized older adults experience higher rates of surgical or treatment-related complications, falls, delirium, infection, pressure ulcers, and dehydration than hospitalized younger adults, with detrimental effects described as cascade iatrogenesis (Thornlow, Anderson, & Oddone, 2009). Under any circumstance, hospitalization is a sentinel event that is associated with many intrinsic, extrinsic, and system-related stressors with significant life-threatening consequences that can be more serious in the context of frailty. In acute-care hospitals, the prevalence of frailty in medical inpatients is high, ranging from 17.9% to 66.4% (Hilmer et al., 2009; Joosten, Demuynck, Detroyer, & Milisen, 2014; Oo, Tencheva, Khalid, Chan, & Ho, 2013; Pilotto et al., 2008; Wou et al., 2013). Consequences of frailty in hospitalized patients include new and worsening morbidity, dependence, disability, early and multiple rehospitalizations, institutionalization, and excess mortality. These outcomes have significant implications for individuals and families, health and social service organizations, and society and are therefore of concern for clinicians and other providers and policy makers (Lafont et al., 2011). As awareness of frailty and its consequences has increased, so has interest in implementing frailty assessment as a way to augment and enhance the accuracy of traditional risk- and disease-focused assessments to improve identification of those individuals who may be more likely to “crash and burn” though they may appear stable and capable of withstanding the stresses of procedures, treatments, and surgery.

There is no gold standard frailty assessment tool for use in the acutely ill hospitalized population. In acute care, clinicians often perform frailty assessment at the foot of the bed (frailogram), basing it on visual inspection and subjective judgment; however, this approach is unreliable for identifying high-risk patients (Hii, Lainchbury, & Bridgman, 2015). Thus, efforts to quantify frailty in this population have intensified in recent years. There are many frameworks for frailty assessment; the most widely cited include the frailty physical phenotype, deficit accumulation, and multidimensional assessment. The frailty physical phenotype defines frailty as a biologic syndrome and specifies five criteria for categorizing level of frailty (i.e., involuntary weight loss, fatigue/exhaustion, slow gait speed, weak handgrip strength, and sedentary behavior; Fried, Ferrucci, Darer, Williamson, & Anderson, 2004). In the deficit accumulation framework, frailty is the cumulative and aggregated burden of disease, disability, symptoms, and cognitive dysfunction operationalized in a frailty index of 30−70 impairments (Theou, Brothers, Mitnitski, & Rockwood, 2013). Multidimensional approaches assimilate impairments derived from the comprehensive geriatric assessment and include physical and cognitive performance tests (Gobbens, van Assen, Luijkx, & Schols, 2012). These frameworks apply different theoretical perspectives, characterize frailty differently, yield widely varying prevalence rates, and identify subgroups with differential risk for negative outcomes (Cigolle, Ofstedal, Tian, & Blaum, 2009; Theou et al., 2013). However, these assessments are difficult to implement in practice because they involve additional data collection, can be time-consuming and burdensome to patients and clinicians, and, in some cases, lack clinical relevance for acute and post-hospital care management (Forti et al., 2014).

The need to assess frailty is undisputed, but valid tools for doing so that meet the needs of busy clinicians and benefit patients are currently lacking (van Kempen, Melis, Perry, Schers, & Rikkert, 2015). The ideal frailty assessment would use clinical data that had already been collected as a part of routine practice, be practical and clinically relevant, accurately identify high-risk patients, minimize clinician and patient burden, and facilitate clinical decision-making and care management. To that end, there is a vast repository of clinical data in the patient electronic health record (EHR) that researchers and clinicians could use in frailty assessment to aid in risk stratification and characterization of frailty.

The purpose of this study, then, was to (1) characterize frailty in hospitalized older adults with a frailty risk score (FRS) using a biopsychosocial framework relying on clinical data from the EHR and (2) determine associations between frailty and time to in-hospital mortality and 30-day rehospitalization.

Method

Design

For this retrospective, cross-sectional, correlational study, we used a convenience sample of hospitalized adults 55 years of age or older admitted from June 1, 2010, through August 31, 2011, to medicine units at a 938-bed, not-for-profit academic-affiliated, tertiary-care hospital located in the southeastern United States.

Sample and Setting

We first constructed an initial sample using a proprietary data query tool (Horvath et al., 2011) to identify all hospital admissions during the study time frame and then delimited the sample based on inclusion and exclusion criteria. Inclusion criteria were (a) 55 years of age or older; (b) admission to general medicine, cardiology, or orthopedic service; (c) overnight hospital stay; and (d) complete data for four serum biomarkers: albumin, C-reactive protein (CRP) or high-sensitivity (hs-) CRP, hemoglobin, and white blood cells (WBCs). We established the age parameter at 55 years or older because emerging research indicates that frailty is not unique to older persons but has been demonstrated in middle-aged adults, especially members of minority racial or ethnic groups (Bandeen-Roche et al., 2015; Hirsch et al., 2006), and determining risk through the lens of frailty in middle-aged and older patients would indicate its potential use in risk stratification. The initial query yielded 690 patients; of these, 281 had data for the four biomarkers. Exclusion criterion was diagnosis of cancer with active treatment. After exclusions, the final sample was 278 independent admissions. We determined that we would be able to detect a small-to-moderate effect size (Cohen’s f ² = .10) in a multiple linear regression with 20 independent variables with sufficient statistical power (90%) with a sample size of 278 patients (Faul, Erdfelder, Buchner, & Lang, 2009).

Ethical Considerations

The institutional review boards of Duke University Hospital and the University of North Carolina at Greensboro approved study procedures.

Data Collection

The first author (D.L.) and two registered nurses trained in study procedures abstracted study data via manual search of clinical data in the EHR, which consisted of electronic PDF files of structured and unstructured information. D.L. systematically cross-checked all abstracted data twice against the EHR for accuracy and completeness, and the second author (D.W.) randomly sampled the data once. They found discrepancies in <1% of the data and reconciled these.

Conceptual Model

We determined potential study variables using a biopsychosocial model and concepts pertaining to inflammaging. The biopsychosocial model is grounded in systems theory and identifies key domains (i.e., biological, psychological, and social), how these domains are interrelated, and the inseparability of the parts of a person’s experience from the whole person (Engel, 1981). The biopsychosocial model has relevance for the multifactorial etiology and complexity of frailty (Cesari, 2011; N. M. de Vries et al., 2011; Gobbens et al., 2012). Inflammaging is the low-grade chronic systemic inflammation and progressive increase in proinflammatory markers that is associated with aging and frailty (Cevenini, Monti, & Franceschi, 2013; Franceschi & Campisi, 2014; Hunt, Walsh, Voegeli, & Roberts, 2010). Inflammaging has damaging effects on cellular and organ function in contrast to acute, transient inflammation, which is a basic immune response to harmful conditions such as infection or trauma but does no lasting damage. Disrupted inflammatory pathways and elevated inflammatory markers are considered features of frailty even if abnormal levels can be explained by disease or nutrition deficiency (Cevenini et al., 2013).

Frailty Risk Score (FRS)

We assessed a pool of potential frailty risk factors gleaned from prior research, consensus opinion, and practice guidelines for their associations with frailty (Andrew, Mitnitski, & Rockwood, 2008; Cevenini et al., 2013; De Martinis, Franceschi, Monti, & Ginaldi, 2006; Fried et al., 2004; Gobbens et al., 2012; Iwata, Kuzuya, Kitagawa, & Iguchi, 2006), and the biopsychosocial model and their availability in the EHR. An expert panel consisting of a board-certified gerontology nurse, geriatrician with expertise in frailty, two advanced practice geriatric nurses, and two doctorally prepared nurses with geriatric expertise provided content validity.

We operationalized frailty by the presence or absence of 16 biopsychosocial risk factors drawn from evidence in previous studies to create an FRS (see Table 1). We further defined six of these risk factors by subfactors. The biological risk factors comprised eight symptoms, syndromes, and conditions and four serum biomarkers. Symptoms, syndromes, and conditions included fatigue, weakness, dyspnea, chronic pain, falls (history or admission diagnosis), vision impairment (glaucoma, cataracts, macular degeneration, retinopathy, blindness), urinary incontinence, and nutrition issues (low body mass index, unplanned weight loss, poor appetite). We selected the four biomarkers—CRP, albumin, hemoglobin, and WBC count—based on associations with frailty, availability in the EHR, and common use in practice. CRP, an acute-phase reactant, exerts catabolic effects leading to muscle atrophy, weakness, fatigue, and poor physical performance due to upregulated protein synthesis and decreased synthesis of albumin. Low albumin and hemoglobin are well-established markers of inflammation and frailty that have similar impacts on symptoms and function. Elevated WBC count is associated with inflammation and frailty and has synergistic interactions with CRP. The three psychological risk factors we included were cognition problems (delirium, dementia), depression, and smoking (current). Finally, we included one social support risk factor (single, living alone, caregiver concerns [i.e., concerns about the impact of illness and hospitalization on discharge needs and planning] or being older, disabled, and living alone). We did not include impaired physical function in the score due to perspectives that regard physical function as an outcome of frailty (Sternberg et al., 2011) and the lack of a valid proxy indicator in the EHR.

Table 1.

Factors Comprising the Frailty Risk Score, Prevalence on Admission, Indicators, and Evidence.

Risk Factor	Present at Admission, n (%)	Indicator Variables	Evidence (selected)
Nutrition issues	222 (79.9)	Any of the following: Weight loss of ≥15 lb over past 6 months; any unplanned weight loss; poor appetite; BMI <18.5 or >30 kg/m2	Bonnefoy et al. (2015); Cooper et al. (2012); Dent, Visvanathan, and Chapman (2012); Goisser, Guyonnet, and Volkert (2016); Jeejeebhoy (2012)
Falls	80 (28.8)	Fall admission diagnosis; history of falls	O. J. de Vries, Peeters, Lips, and Deeg (2013); Joosten et al. (2014); Kojima (2015); Samper-Ternent, Karmarkar, Graham, Reistetter, and Ottenbacher (2012); Zhu et al. (2016b)
Weakness	227 (81.7)	Weakness	Forti et al. (2014); Manty et al. (2012); Stiffler, Finley, Midha, and Wilber, 2013; Toye, White, & Rooksby, 2006
Vision impairmenta	100 (36.0)	Any of the following: ICD-9/10 diagnosis of cataracts (corrected, uncorrected), glaucoma, macular degeneration, retinopathy, blind	Cesari, et al., 2004; Chen, Yen, Dai, Wang, and Huang (2011); Klein, Klein, Knudtson, and Lee (2003, 2005); Knudtson, Klein, and Klein (2009)
Dyspnea	104 (37.4)	Dyspnea or shortness of breath	Brinkley et al. (2009); Galizia et al. (2011); Lahousse et al. (2016); Vaz Fragoso, Enright, McAvay, Van Ness, and Gill (2012)
Fatigue	249 (89.6)	Any of the following: fatigue, sleep problems, CPAP at night	Fried et al. (2004); Jeejeebhoy (2012); Vaz Fragoso, Gahbauer, Van Ness, and Gill (2009)
Chronic pain	173 (62.2)	Chronic pain	Blyth et al. (2008); Shega et al. (2012); Wade et al. (2016)
Urinary incontinence	66 (23.7)	Urinary incontinence	Berardelli et al. (2013); DuBeau, Kuchel, Johnson, Palmer, and Wagg (2010); Miles et al. (2001)
Smoking	45 (16.2)	Current smoker	Hubbard et al. (2009); Kojima (2015)
Depression	97 (34.9)	ICD-9/10 diagnosis of depression	M. H. Artz et al. (2015); Lohman, Dumenci, and Mezuk (2016); Mezuk et al. (2012); Vaughan, Corbin, and Goveas (2015)
Cognition problems	88 (31.7)	Any of the following: ICD-9/10 diagnosis of dementia, any type; Alzheimer’s disease; any memory problems; abnormal mental status/delirium	Ávila-Funes et al. (2011); Han, Lee, and Kim (2014); Kelaiditi et al. (2013); Quinlan et al. (2011); Panza et al. (2011); Robertson, Savva, and Kenny (2013)
Social support issues	176 (63.3)	Any of the following: living alone; single; caregiver concerns; older adult, disabled, and live alone; clergy visit	Andrew et al. (2008); Bilotta et al. (2010); Chamberlain et al. (2016); Dent et al. (2014); Freiheit et al. (2010)
High CRP or hs-CRP levelb	246 (88.5)	Abnormal flag, high	Chang et al. (2011); Collerton et al. (2012); Fontana et al. (2013); Gale et al. (2013); Hubbard and Woodhouse (2010); Rønning et al. (2010); Sarkisian et al. (2008); Zhu et al. (2016a)
Low albumin levelc	258 (92.8)	Abnormal flag, low, <3.6 g/dl	Arques et al. (2008); Fontana et al. (2013); Forti et al. (2014); Kim, Higgins, Canaday, Burant, and Hornick (2014); Revenig et al. (2015); Smit, Winters-Stone, Loprinzi, Tang, and Crespo (2013)
Low hemoglobin leveld	267 (96.0)	Abnormal flag, low (women <12.0 g/dl, men <13.7 g/dl	A. S. Artz (2008); Fontana et al. (2013); Hirani et al. (2015); Haslam et al. (2012); Khandelwal et al. (2012); Pires Corona, Drumond Andrade, de Oliveira Duarte, and Lebrao (2015); Röhrig (2016)
Low or high WBC counte	206 (74.1)	Abnormal flag, high or low, <3.2 or >9.8 × 109/mcl	Barzilay et al. (2007); Baylis et al. (2013); De Martinis et al. (2006); Leng et al. (2009); Li et al. (2011); Yao et al. (2011)

Note. N = 278. All risk factors were measured via self-report and/or nursing or physician assessment in the EHR. Laboratory assay equipment used: Albumin (Beckman Coulter Unicel DXC 600-800, Bromocresol Purple [dcp] time test endpoint); CRP and hs-CRP (Beckman Synchron Systems Neon Infrared Particle immunoassay rate methodology); WBC and hemoglobin (electronic impedance differential lysis fluorescent flow cytometry and colorimetric measurement). CRP = C-reactive protein; EHR = electronic health record; hs-CRP = high-sensitivity C-reactive protein; WBC = white blood cell; ICD = international classification of disease; CPAP = continuous positive airway pressure.

^aVision impairment was measured via self-report, nursing or physician assessment, and/or ophthalmology clinic note in the EHR. ^bReference range: >0.6 pg/dl for CRP or >0.30 mg/dl for hs-CRP. ^cReference range: 3.5−4.8 g/dl. ^dReference range: women 12.0−15.5 g/dl; men 13.7−17.3 g/dl. ^eReference range: 3.2−9.8 × 10⁹/mcl.

To calculate the FRS, we counted each risk factor as yes = 1 if present and no = 0 if not present. For the six risk factors further defined by more than one subfactor (i.e., nutrition issues, falls, vision impairment, fatigue, cognitive problems, social support issues), the presence of at least one subfactor resulted in counting the overall risk factor as present (see Table 1). The biomarker risk factors were operationalized by the categorical abnormal flag, which indicated that the laboratory value fell outside the reference range, high or low. We created an FRS as the unweighted count of risk factors present (theoretical range = 0−16), where higher scores are indicative of increased frailty. We then used these FRSs to model the outcome variables of time to in-hospital mortality and rehospitalization within 30 days of discharge.

Other Variables

Other variables for sample description that we included in the analyses were age, gender, partner status, race/ethnicity, impaired activities of daily living (ADL), comorbidity, medication count, preadmission location, living arrangement, and discharge location. Comorbidity was the count of diagnoses listed on admission. Medication count was the number of routine prescription and nonprescription medications. All data reflected each patient’s health status on admission.

Data Analysis

Demographic and clinical characteristics are reported as a percentage of the total sample or as the mean (M) ± standard deviation (SD). The only missing data were for one case on one risk factor. Time to in-hospital death was defined as days between admission and discharge, where a patient was considered censored if still alive at discharge. Extended Cox regression modeling (Therneau & Grambsch, 2000) was performed, as there were no proportional hazards for FRS (p < .001). Time-specific hazard ratios (HRs) were estimated using the HAZARDRATIO command in the PROC PHREG SAS procedure, where times were chosen corresponding to the 10th, 25th, 50th, 75th, and 90th percentiles for those with and without events.

Logistic regression was performed for 30-day rehospitalization for persons still alive at discharge (n = 265). All modeling was adjusted for age, gender, and race. Linearity of associations was checked with visual inspection of scatterplots with locally weighted linear regression smoother curves and higher order polynomial terms after centering. No substantial nonlinearity was found in these analyses. Goodness of fit for logistic models was checked with Hosmer–Lemeshow testing. Cutpoints on the FRS indicative of increased risk for outcomes were estimated using recursive partitioning to facilitate decision rule making (Stiell & Wells, 1999; Therneau, Atkinson, Ripley, & Ripley, 2015). Analyses were performed using SPSS (IBM SPSS Statistics for Windows, Version 21.0, IBM Corp., Armonk, NY), SAS (SAS Institute Inc., Cary, NC), and Mplus (Muthén & Muthén, 2015). A two-sided p value < .05 was considered statistically significant.

Results

The sample included 278 hospitalized medical patients who were 55−98 years of age (M = 70.2 years, SD = 10.3). We have summarized sample characteristics in Table 2. Over half were female (53%), one third were non-White (36%), about half were single, divorced, or widowed (49%), and approximately one fifth lived alone. Most discharges were to home (54%), and 13 patients (4.7%) died during hospitalization. Thirty-three patients (11.9%) were rehospitalized within 30 days of discharge. The mean length of stay (LOS) was 9.92 days (SD = 9.58, range = 1−72). LOS was 4 days at the 25th percentile, 7 days at the 50th percentile, and 12 days at the 75th percentile. The average FRS was 9.4 (SD = 2.2, range = 2−15), with 4.3% having two to five risk factors and 17.3% having 12 to 15 risk factors.

Table 2.

Demographic and Clinical Characteristics of the Hospitalized Older Adults.

Variable	n (%) or M (SD; range)
Age (years), mean (SD; range)	70.2 (10.3; 55−98)
Sex, female, n (%)	146 (52.5)
Race/ethnicity, n (%)
White (Caucasian)	178 (64.0)
Black (African American)	92 (33.1)
American Indian, Asian, other	8 (2.8)
Service, n (%)
General medicine	194 (69.8)
Cardiology	55 (19.8)
Orthopedic	29 (10.4)
Partner status, n (%)
Married	142 (51.1)
Single, divorced, widowed	136 (48.9)
Admission location, n (%)
Home	224 (80.6)
Other (extended care, rehabilitation facility, other hospital)	54 (15.4)
Discharge, n (%)
Home	150 (54.0)
Skilled nursing, rehabilitation, another hospital	105 (37.8)
Hospice (home or facility)	10 (3.6)
In-hospital mortality	13 (4.7)
Occupation, n (%)
Employed	28 (10.2)
Retired	179 (64.4)
Disabled	58 (20.9)
Unemployed	13 (4.7)
Diagnoses, n (%)
Hypertension	229 (82.4)
Angina	55 (19.8)
MI	69 (24.8)
Heart failure	91 (32.7)
Stroke	57 (20.5)
Diabetes mellitus	136 (48.9)
COPD	70 (25.2)
Osteoarthritis	137 (49.3)
CKD	103 (37.1)
Cancer	63 (22.7)
BMI (kg/m2), mean (SD; range)	28.6 (7.6; 13−65)
Comorbidity (count of ICD 9/10 diagnoses), mean (SD; range)	13 (4.6; 1−26)
Medications (routine prescription or nonprescription), mean (SD; range)	11.9 (5.2; 0−31)
Frailty risk score, mean (SD; range)	9.4 (2.2; 2−15)
Length of stay (days), mean (SD; range)	9.9 (9.6; 1−72)
Hospitalizations during 15-month study period, mean (SD; range)	1.6 (0.9; 1−8)
Rehospitalization within 30 days of discharge, n (%)	33 (11.9)

Note. N = 278. All measures were at hospital admission except for length of stay and rehospitalization. BMI = body mass index; CKD = chronic kidney disease; COPD = chronic obstructive pulmonary disease; MI = myocardial infarction.

The instantaneous risk of in-hospital death depended upon follow-up time after admission (i.e., number of days of hospitalization from admission to death, p < .001). Figure 1 provides estimated HRs of FRS effects on time to in-hospital death for select percentiles of follow-up time. Increased FRSs are associated with increased instantaneous risk of in-hospital death approximately around 3 days ≤ LOS ≤ 7 days and then become nonsignificant until extreme LOSs, where the association flips direction. For example, at 3 days postadmission, each additional 1-point increase in the FRS is associated with a 127% increase in the instantaneous risk of in-hospital death (adjusted HR = 2.27, 95% confidence interval [CI] = [1.40, 3.67]), controlling for age, race/ethnicity, and gender. However, at 18 days, the effect of the FRS is null (adjusted HR = 0.89, 95% CI = [0.55, 1.46]). This finding might imply that there is an initial window of high risk of in-hospital mortality after admission where increased frailty is most predictive. To this end, Figure 2 presents estimated cutoffs of FRSs based on decision trees from recursive partitioning. Scores ≥10 are most salient for in-hospital mortality if the patient is female, <80 years of age, and White. Based on a cut-off score of 9, the prevalence of frailty was 68%. Compared to the non-frail, frail patients were older (M = 71.6 years ± 10.5 vs. M = 67.2 years ± 9.3), female (56% vs. 47%), and non-White (39% vs. 31%).

Figure 1.

Hazard ratios (HRs) of the frailty risk score from extended Cox modeling of time to in-hospital death (N = 278). Adjusted HRs are estimated at the 10th, 25th, 50th, 75th, and 90th percentiles of follow-up time according to event status (alive or died), controlling for age, race/ethnicity, and gender. Error bars are 95% confidence intervals for the adjusted HR.

Figure 2.

Decision tree for time to in-hospital mortality from recursive partitioning (N = 278). Non-White versus White ≥0.5 = Non-White in the tree. Female versus male ≥0.5 = female.

Higher FRSs were only marginally associated with increased odds of 30-day rehospitalization in multivariable logistic regression models among patients who did not die in hospital (adjusted odds ratio = 1.18, 95% CI = [0.98, 1.43], p = .086). The area under the receiver operating characteristic (ROC) curve (AUC) was significantly above 0.5 (AUC = 0.66, 95% CI = [0.57, 0.76], p = .003). Figure 3 provides the ROC curve for 30-day rehospitalization. Based on a decision tree from recursive partitioning, a cut-off score of ≥9 is most salient for rehospitalization among White patients who were alive at discharge (see Figure 4).

Figure 3.

Receiver operating characteristic (ROC) curve for 30-day rehospitalization. Curve based on multivariable logistic regression with frailty risk score, age, race/ethnicity, and gender for patients who did not die in hospital (n = 265). Area under the ROC curve = 0.662 (p = .003, 95% confidence interval = [0.568, 0.755]).

Figure 4.

Decision tree for 30-day rehospitalization from recursive partitioning. Based on patients who did not die in hospital (n = 265). Non-White versus White ≥0.5 = Non-White in the tree. Female versus Male ≥0.5 = Female.

Finally, as we had created the FRSs from an unweighted count that treated each of the 16 risk factors as equally important, we performed post hoc sensitivity analyses to examine this inherent assumption by creating weighted FRSs with weights based on estimated factor loadings of a one-factor confirmatory factor analysis model for categorical indicators (Brown, 2015). Modeling of outcomes with this weighted FRS yielded results for time to in-hospital death and 30-day rehospitalization that were consistent with the results for the unweighted score.

Discussion

In the present study, we examined frailty in acutely ill hospitalized adults ≥55 years of age using readily available clinical data in the EHR that included symptoms, syndromes, conditions, and serum biomarkers. Our findings provide content validity for select frailty risk factors and predictive validity for an FRS in the context of study outcomes of time to in-hospital mortality and 30-day rehospitalization.

Frailty assessment is important in acute-care hospitals since frailty, though common, may not be apparent to clinicians, family members, and others. Growing evidence suggests that the prevalence of frailty among hospitalized older adults is as high as 50−94%, depending on the instrument used (Basic & Shanley, 2015; Dent, Chapman, Howell, Piantadosi, & Visvanathan, 2014; Dent & Hoogendijk, 2014; Forti et al., 2014; Oo et al., 2013; Theou et al., 2013), and its deleterious consequences put an increased burden on patients, caregivers, and the healthcare system (Basic & Shanley, 2015; Joosten et al., 2014). In the present study, the prevalence of frailty was 68% based on a cut-off FRS of 9. Given that the patients comprising our sample were hospitalized with acute illness, we would expect the prevalence of frailty to be somewhat higher than that of the general population of adults aged ≥55 years, and our findings must be considered in this context.

The frailty risk score we developed in this study is a measure of risk or vulnerability designed to reflect biopsychosocial and physiological complexity, medical acuity, and symptom burden in a way that individual assessments of symptoms, syndromes, diseases, and even older age may not capture. For example, commonly used risk assessment tools for falls, delirium, or pressure ulcers identify risk level for a particular clinical concern, whereas the FRS may provide a measure of aggregated risk. In addition, while age is often used as a primary risk indicator, in the present study, while it was significantly associated with the FRS (β = .039, t = 3.16, p = .002) in additional bivariate linear regression analyses, it was no longer significant in multiple regression models controlling for other variables. Further post hoc analyses suggested that the effects of the FRS on the study outcomes of time to in-hospital mortality and 30-day rehospitalization were not moderated by age (all p > .10). These findings correspond with those of prior studies (Basic & Shanley, 2015; Bellal et al., 2014; Santos-Eggimann, Cuénoud, Spagnoli, & Junod, 2009; Szanton, Allen, Seplaki, Bandeen-Roche, & Fried, 2009) and demonstrate that frailty is distinctly different from age and age-related changes that develop overtime in non-frail persons (Mansur, Colugnati, Grinsenkov, & Bastos, 2014). Further, the weak effect size we found between age and FRSs challenges the notion that frailty is unique to the aged. Although most prior research has identified a significant association between older age and frailty, recent studies (cited above) have reported a lack of association and a notable prevalence (36% prefrail and 1.3% frail) of frailty in a large population-based study of middle-aged adults (Guessous et al., 2014). A better understanding of the construct of frailty should lead to an increased appreciation of the heterogeneity seen in aging relative to functional decline and other outcomes (Bergman et al., 2007). It is also important to note that frailty assessment tools that include older age (e.g., >75 years) as a screening criterion would fail to detect frailty in vulnerable younger individuals.

The frailty risk score significantly predicted in-hospital mortality in medical inpatients, with higher risk of in-hospital mortality associated with higher scores, after controlling for age, gender, and race. The prevalence of in-hospital mortality in our sample was 4.7% (n = 13), similar to that found in prior research on frailty and patient outcomes (Basic & Shanley, 2015; Evans, Sayers, Mitnitski, & Rockwood, 2014). Other investigators have reported that frailty predicts early mortality (Dramé et al., 2008) and mortality up to 1 year in medical inpatients (Evans et al., 2014; Iwata et al., 2006; Joosten et al., 2014; Khandelwal et al., 2012; Pilotto et al., 2012). Though there was an association between FRS and in-hospital mortality in the present study, mortality actually may not have been caused by frailty. As Howlett and colleagues previously noted, “not everyone dies a frail death” (Howlett, Rockwood, Mitnitski, & Rockwood, 2014, p. 6). Rather, mortality could have been due to disease states or even iatrogenic factors associated with hospitalization. Investigation into cause of death was, however, outside the scope of this study. Interestingly, though prior research has found in-hospital mortality to be associated with longer LOS (Basic & Shanley, 2015), in the present study we found that the highest mortality risk occurred during the first 3 days of hospitalization and those who died during that period had higher FRSs. Thus, greater awareness of the adverse impact of frailty and the increased risk of mortality during these early days of hospitalization may alert clinicians to intensify medical and nursing monitoring and treatment. As a tool for risk appraisal, the FRS may serve to alert clinicians to a more vulnerable health status.

Surprisingly, we found that the weighted and unweighted FRSs yielded similar results, suggesting that there were no strong individual drivers of frailty among the factors. This finding may facilitate ease of use of the tool by clinicians since computing an unweighted score is additive and does not involve complex computations. The frailty risk factors verified as present in individual frailty scores could guide clinical decision-making and tailored care management. In prior research examining assessment tools similar to the FRS, symptom groupings had stronger associations with mobility dysfunction than did disease scores. For example, in one study, weakness, pain, and shortness of breath comprised the most common symptom typology associated with mobility dysfunction (Whitson et al., 2009). In research investigating frailty subtypes using multiple indicators (nutrition, physical activity, mobility, strength, energy, cognition, mood, interleukin-6 [IL-6], CRP, the five frailty phenotype markers), researchers similarly identified several unique clusters of risk factors and biomarkers (Sarkisian, Gruenewald, Boscardin, & Seeman, 2008; Sourial et al., 2010). Future research on the FRS can assess for significant clusters or subgroups of the risk factors.

In the present study, the 30-day rehospitalization rate was lower than recent estimates for the U.S. average for noninstitutionalized Medicare beneficiaries: 11.9% versus 17.3% (Gorina, Pratt, Kramarow, & Elgaddal, 2015), respectively, and the FRS was only marginally associated with rehospitalization. However, we did not investigate readmission at other hospitals or rehospitalization beyond 30 days; therefore, our findings likely underestimated rehospitalization. Notable among our findings, though, was that more than 60% of subjects had been hospitalized at least once during the 15 month study time frame: of these, 37% had ≥2 hospitalizations and 4% had ≥4 hospitalizations. Wou et al. (2013) found that, among 667 medical inpatients, rehospitalization within 90 days was the most common adverse outcome, affecting 26.2%. In the United States, frequent rehospitalization (≥3/year) is a major concern, with older adults among those at greatest risk (Gorina et al., 2015). Rehospitalization has been associated with greater medical acuity and clinical ambiguity, inadequate social support, ill-prepared caregivers, and lack of community resources, and the presence of frailty may be another important risk factor (Hunt, Walsh, Voegeli, & Roberts, 2013). Reducing hospitalizations is a health-care priority. Many are avoidable, and early rehospitalization is considered to be a failure of clinical and organizational systems that ensure quality of care and safety during transitions. Reducing rehospitalizations may thus require intensified discharge planning. Future research should target frequently readmitted patients and investigate factors associated with both early and later rehospitalization since different interventions may be more effective in the two instances (Hansen, Young, Hinami, Leung, & Williams, 2011; Leppin, Gionfriddo, Kessler, Brito, & Mair, 2014; van Walraven, Jennings, & Forster, 2012). Frailty assessment may facilitate these efforts by helping clinicians and researchers to target specific risk factors, especially symptoms and psychosocial issues, such as caregiver concerns and social support, that are pivotal in many rehospitalizations (Hunt et al., 2013).

Among patients in the present sample, there was a high prevalence of symptoms such as fatigue, weakness, chronic pain, and dyspnea. An important feature of frailty, however, is that not all frail persons experience the same symptoms, and frailty can exist in the absence of a specific disease (N. M. de Vries et al., 2011). The FRS may capture the effects of acute and chronic illness, abnormal physiologic parameters, and symptom burden better than medical diagnoses and may signal the need for interventions that address more than one diagnosis and/or symptom at a time. Frailty assessment using such a tool may not only help identify high-risk patients who may require more intensive personalized intervention for multiple risk factors, it may also help identify patients at lower risk of adverse effects from medical or surgical procedures (Wou et al., 2013). Hospitals are increasingly employing standardized risk assessment tools for falls, delirium, nutrition, and skin integrity (pressure ulcers) to improve patient care. The use of an FRS to perform a single assessment that incorporates all of these measures in addition to others may enhance precision and validity, better characterize frailty, and allow clinicians to identify higher priority, potentially interrelated risk factors.

Research has yet to identify the most important biologic indicators and physiologic markers for frailty (Fernández-Garrido, Ruiz-Ros, Buigues, Navarro-Martinez, & Cauli, 2014; Zaslavsky et al., 2013), especially biomarkers that would be applicable in acutely ill hospitalized patients. In the present study, we selected the inflammatory biomarkers hemoglobin, albumin, CRP, and WBC as factors in the FRS because they are commonly used in practice. From a practical standpoint, adding costly laboratory tests in acute care for frailty assessment may be neither feasible nor justifiable; additional laboratory testing potentially increases health-care costs and patient burden and may not contribute meaningfully to individual care management. Physiologic measures that augment clinical impressions based on observation and judgment, patient self-report, and performance tests are an important part of frailty assessment. Frailty’s association with a pro-inflammatory state as well as the complexity of the syndrome of frailty and the multiple overlapping pathophysiologic mechanisms involved all lend importance to the use of biomarkers as part of this assessment (Chang, Vaz Fragoso, Van Ness, Fried, & Tinetti, 2011; Leng, Chaves, Koenig, & Walston, 2002; Yao et al., 2011). Researchers have used various biomarkers to assess low-grade inflammation, such as CRP, IL-6, tumor necrosis factor (TNF), WBC, hematologic and coagulation factors, and stress hormones (Compté, Bailly, De Breucker, Goriely, & Pepersack, 2015; Puts, Visser, Twisk, Deeg, & Lips, 2005; Walston et al., 2002). Hemoglobin, albumin, CRP, and WBC, in particular, are integrally linked to the immune system and the inflammatory response in frailty (Haslam et al., 2012; Leng et al., 2002; Li, Manwani, & Leng, 2011; Wu, Shiesh, Kuo, & Lin, 2009). In the present study, the majority of patients had abnormal levels of these four biomarkers, but interpretation is difficult since acute illness may confound biomarker levels. Albumin and hemoglobin levels (in the absence of hemorrhage) are less dramatically affected by acute illness, and low levels more likely reflect baseline status. Elevated CRP and WBC levels, however, are more likely associated with the acute illness. Our use of the categorical abnormal flag to indicate the presence or absence of a frailty risk factor may have contributed to a ceiling effect and limited the usefulness of these factors for risk prediction. The use of the actual laboratory value interpreted as a continuous variable might provide for more nuanced interpretations of within-range or borderline scores. The fact that a majority of patients had abnormal flags did not allow for the determination of subrisk groups based on the degree to which lab values differed from the normal range. Exploring the use of these biomarkers as continuous variables may yield different outcomes in future research. Additionally, our requirement that the sample consist of patients who had all four biomarkers present in the EHR may have differentiated this sample from hospitalized patients for whom these measures have not been ordered for clinical reasons.

Prior research on the use of inflammatory biomarkers in frailty assessment is limited, and results are mixed regarding evidence supporting the inclusion of these biomarkers in an FRS, especially when applied in hospitalized older adults. In a recent systematic review, Fernández-Garrido, Ruiz-Ros, Buigues, Navarro-Martinez, and Cauli (2014) found levels of the inflammatory biomarkers CRP, IL-6, and TNF to have strong associations with incident and prevalent frailty. In a large longitudinal study in community living adults 60–90 years, high concentrations of CRP and fibrinogen were predictive of incident frailty in women but not in men (Gale, Baylis, Cooper, & Sayer, 2013). Other researchers, however, found no associations when classifying frailty using different assessment tools (Hubbard, Searle, Mitnitski, & Rockwood, 2009; Leng, Xue, Tian, Walston, & Fried, 2007). In cross-sectional population-based studies, higher CRP level was associated with frailty (Collerton et al., 2012) and overnight hospitalization (Zhu et al., 2016b). In a study among hospitalized older patients, higher level of hs-CRP predicted mortality within 2 months of discharge, though prealbumin level did not (Nouvenne et al., 2016). Studies have also demonstrated that abnormal WBC count is associated with frailty in community-dwelling older adults (Fernández-Garrido et al., 2014; Walston et al., 2002) including the Women’s Health and Aging Studies conducted among disabled older women (Leng et al., 2009). Using a comprehensive geriatric assessment to assess frailty, researchers found that, of the 17 serum biomarkers examined, albumin, CRP, hemoglobin, WBC, glucose, erythrocyte sedimentation rate, cholesterol, insulin-like growth factor-1, and triiodothyronine were significantly associated with frailty and mortality (Fontana et al., 2013). Findings of strong associations between low hemoglobin, anemia (A. S. Artz, 2008; Röhrig, 2016; Roy, 2011), and low albumin and frailty indicators such as fatigue, weakness, and decreased physical and cognitive function (Hazzard, 2001; Robinson et al., 2011; Wu et al., 2009) make these biomarkers attractive for use in frailty assessment. Thus, although no one specific biomarker has been identified that will definitively predict frailty, research has shown that aberrations in biomarkers are significantly associated with frailty and suggests that the use of multiple biomarkers may better convey a physiologic signature that reflects the multisystem dysregulation of frailty (Calvani et al., 2015; Cevenini et al., 2013; Fontana et al., 2013; Franceschi & Campisi, 2014). There are ongoing efforts to identify biomarkers that may improve the differential diagnosis of frailty, to determine whether clusters of biomarkers distinguish different forms of frailty, and to use biomarkers to facilitate monitoring of the progression or regression of the symptoms of frailty and responses to interventions (Erusalimsky et al., 2016). For example, researchers used a frailty index comprising 21 routine laboratory tests plus systolic and diastolic blood pressure to characterize older adults living in the community (Howlett et al., 2014) and in nursing homes (Rockwood, Mitnitski, & Howlett, 2015) by frailty level and found that as frailty scores increased, so did the risk of 6-year mortality. Most of the studies on associations between biomarkers and frailty have been among community-dwelling older adults rather than acutely ill hospitalized older patients. The present study is among only a few to explore these associations among hospitalized older adults.

One concept that may capture the multisystem physiologic dysregulation in frailty and help characterize the relationship between biomarkers and frailty is that of allostatic load (Beckie, 2012). Allostatic load could be particularly useful in exploring the underlying mechanisms for failing homeostasis and compensatory responses to stressors in the progression and natural history of frailty (Kuchel, 2009; Zaslavsky et al., 2013). Small changes in individual biomarkers (borderline high or low) may exert negligible impact on systems (because of functional redundancies) but, when these multiple changes are aggregated, the magnitude of the effect is much greater. For example, in a large community-based cohort study, investigators assessed allostatic load using 13 biomarkers for cardiovascular, endocrine, immune, and metabolic function (included CRP, hemoglobin, systolic and diastolic blood pressure) and found that, in adjusted models, a 1-unit increase in allostatic load at baseline was associated with a 10% greater likelihood of frailty at 3-year follow-up (Gruenewald, Seeman, Karlamangla, & Sarkisian, 2009). Similarly, in the Women’s Health and Aging Study, in which researchers assessed allostatic load using 11 biomarkers and diastolic and systolic blood pressure, higher allostatic load was associated with increasing odds for frailty (Szanton et al., 2009). These findings support the premise that multisystem physiological dysregulation manifested in biomarker aberrations may represent physiologic responses to changes associated with aging as well as to psychosocial and environmental demands and chronic diseases and may be a warning sign for frailty. The notion that the sum is greater than the parts and that small changes can have large effects, as posited in complexity science (Zaslavsky et al., 2013), should direct further research using frameworks such as allostatic load and deficit accumulation (Theou et al., 2013) to refine the FRS.

For the present study, we used data that were stored in an electronic format as part of the EHR; however, the data were not electronically searchable, and the required hand search of electronic files was time-consuming and would be impractical in clinical practice. The increasing adoption of fully integrated EHRs by hospitals will provide access to a substantial repository of data available in a more searchable format using informatics. This development presents important opportunities for more effective and timely use of the extraordinary amount of data that nurses and other care providers routinely collect, data that might otherwise be incomprehensible and not fully appreciated as an aggregated whole (Samuels, McGrath, Fetzer, Mittal, & Bourgoine, 2015). There may be considerable challenges to accessing the data to construct a frailty score in a usable format, thus, information technology expertise will facilitate the development and use of data query tools and algorithms to automatically retrieve clinical data and provide clinical decision-making support and care planning. The opportunity to apply novel data analytics techniques and data visualization also holds promise for improving how information is used and shared in order to inform and improve clinical practice, especially information contained within large data sets. For example, in addition to care planning, the FRS may have utility for assessing patient acuity, since it includes multiple parameters that are relevant to medical complexity and instability with implications for risk prediction and determination of resource needs and utilization. Also, if FRSs were being generated on a regular basis from clinical assessments, they could serve as an early warning to signal significant changes in condition and emerging cascade iatrogenesis (Thornlow et al., 2009). The use of a longitudinal study design as well as automated collection and calculation of FRSs are important next steps for further research to demonstrate validity, reliability, and clinical relevance of this score.

This study had several limitations. The retrospective, cross-sectional design limited our ability to assess frailty to one time point, admission, yet the level of frailty may change during the course of hospitalization. Also, retrospective data collection relies on the accuracy and completeness of nurse and physician documentation, which was originally recorded for clinical and not research purposes. In addition, the study setting was one U.S. hospital, and the population, provider practice patterns, and system characteristics likely differ across settings and locations. Because we characterized frailty on admission in patients hospitalized for acute illness, the health status of our subjects may have reflected the effects of acute illness and other factors that contribute to the vulnerability of frailty, but do not underlie frailty. Finally, we limited study variables to those that were available in the EHR, and some variables may be unique to the study setting. Care providers had ordered the assessment of the four biomarkers we used in the study for all study subjects. Although practitioners may routinely order measurement of several of these biomarkers in acutely ill patients, CRP, for example, may be less common. Also, the timing of biomarker acquisition was not controlled, and baseline data on biomarker levels prior to acute illness were not available. Despite these limitations, the present study provides a potentially useful approach to frailty assessment that captures biopsychosocial and physiologic parameters in a parsimonious tool using existing clinical data in the EHR.

Conclusion

The aging of the population means that the incidence of frailty will continue to increase. It is imperative that health-care professionals not dismiss frailty as an unremitting terminal condition without possibility of improvement or as a reason to withhold care. Instead, care providers should use frailty assessment to optimize person-centered care (Afilalo et al., 2014). Early recognition and prevention of frailty by nursing and other health professionals is crucial to promoting recovery and rapid return to better health and well-being. Interventions such as promoting early mobilization, providing regular toilet assistance, avoiding prolonged food or fluid restriction, offering nutritional support, preventing delirium, promoting safety awareness, and performing optimal symptom management are fundamental approaches to preventing the destabilizing circumstances that may trigger the development frailty and its consequences. In addition, interventions to target chronic systemic inflammation may alleviate or slow the progression of frailty (Gale et al., 2013; Hunt et al., 2010). Finally, evaluation of the hospital environment and practices and improvement in care processes and transitions will also be crucial to preventing the development of frailty.

The frailty assessment we have proposed in the present study is a practical and clinically relevant approach to identifying hospitalized older adults at higher risk of frailty using existing data from the EHR. The FRS provides a composite view of symptoms, syndromes, conditions, and biomarkers that facilitates its use in clinical practice. Further testing of the score in fully integrated EHRs will allow for the refinement of this measure and elucidate its clinical potential in larger and more diverse samples.